The research presents a novel approach to evaluating causal models when randomized controlled trials (RCTs) are not feasible or ethical. The authors propose a new estimator that reduces estimation variance in conditionally randomized experiments based on Inverse Probability Weighting (IPW). This pairs estimator applies the same IPW estimator to both the model and true experimental effects, effectively reducing the variance due to IPW, and achieves near-RCT performance. It paves the way for more robust and reliable model assessments without requiring complex modifications of the IPW estimator.
Publication date: 3 Nov 2023
Project Page: https://arxiv.org/abs/2311.01902v1
Paper: https://arxiv.org/pdf/2311.01902
